Add Count distinct support #741 #1073
Conversation
There was a problem hiding this comment.
The bytecode plan for this differs quite significantly from SQLite -- SQLite uses an ephemeral index and the Found instruction to skip rows where the key is already found in the ephemeral index. This PR's implementation relies on creating an owned string for every value and then conditionally inserting the stringified value to a HashSet.
Moreover, I don't think specialcasing COUNT(DISTINCT foo) is a terribly good idea - what about AVG(DISTINCT foo), SUM(DISTINCT foo) and so on? What about DISTINCT without an aggregation? There should be a single distinctness mechanism and the aggregate function should not be tightly coupled to it.
Unfortunately having such a mechanism depends on us having an OpenEphemeral implementation which we do not yet have.
tests in PR #1066